How to convert JSON to JSON Lines
Updated · 3 min read
Many data platforms want newline-delimited JSON rather than one big array: fine-tuning APIs, BigQuery loads, Elasticsearch bulk imports and streaming tools. This converter writes each element of your JSON array on its own line in compact form, with nothing else in the file.
- Paste a JSON array or drop a
.jsonfile. API responses like{"data": [...]}are unwrapped automatically. - Each element is written on one line without extra whitespace.
- Check the row count to confirm the number of records.
- Download the
.jsonlfile.
Example
JSON input
[
{ "prompt": "Summarize: The meeting moved to Friday.", "completion": "Meeting now Friday." },
{ "prompt": "Summarize: Invoice #42 was paid in full.", "completion": "Invoice 42 paid." }
]JSON Lines output
{"prompt":"Summarize: The meeting moved to Friday.","completion":"Meeting now Friday."}
{"prompt":"Summarize: Invoice #42 was paid in full.","completion":"Invoice 42 paid."}Each object occupies exactly one line, and characters such as non-Latin letters are kept as UTF-8 rather than escaped.
Your data stays on your device. The conversion runs in JavaScript inside this tab — files and pasted text are never uploaded to DataSwitch Pro.
Formatting rules
- One JSON value per line, serialised compactly — no indentation, no trailing commas.
- Line breaks inside strings are escaped as
\n, so each record stays on a single line. - Lines are separated by a single newline, and the file is UTF-8 encoded.
- A wrapper object holding one array (
{"results": […]}) is unwrapped; any other single object becomes one line.
Preparing fine-tuning data
LLM fine-tuning services expect one training example per line, typically an object with a messages array of role/content pairs. Build or clean your examples as a regular JSON array, which is easier to edit, then convert it here. Validate a few lines against the provider's documentation before uploading a large file.
Loading into BigQuery and Elasticsearch
BigQuery loads JSONL with --source_format=NEWLINE_DELIMITED_JSON. Elasticsearch's bulk API also uses NDJSON, but alternates action lines and document lines, so add the action lines (for example {"index":{}}) before each document when preparing a bulk request.
Convert JSON to JSON Lines in code
For one-off files the tool above is quickest. To automate the same task in a script or pipeline, these snippets are a good starting point.
Python
import json
with open("data.json", encoding="utf-8") as f:
rows = json.load(f)
with open("data.jsonl", "w", encoding="utf-8") as f:
for row in rows:
f.write(json.dumps(row, ensure_ascii=False) + "\n")Command line (jq)
jq -c '.[]' data.json > data.jsonlBigQuery load
bq load --source_format=NEWLINE_DELIMITED_JSON --autodetect \
mydataset.mytable data.jsonlFrequently asked questions
What if my JSON is not an array?
An object wrapping a single array is unwrapped; any other object is written as one line, which is still valid JSON Lines.
Can I convert CSV or Excel straight to JSONL?
Yes. Choose CSV or Excel as the source and JSON Lines as the target.
Does it validate my data for fine-tuning?
It guarantees valid JSON on every line, but not the provider-specific structure. Check the required fields in your provider's documentation.
How do I convert back?
Use the swap button or the JSONL to JSON converter.
Is my data uploaded?
No. The conversion runs in your browser.